Bitcoin Forum
October 19, 2019, 08:30:40 AM *
News: 10th anniversary art contest
 
   Home   Help Search Login Register More  
Pages: « 1 [2] 3 4 5 6 7 8 9 10 11 12 13 »  All
  Print  
Author Topic: Logarithmic (non-linear) regression - Bitcoin estimated value  (Read 113562 times)
This is a self-moderated topic. If you do not want to be moderated by the person who started this topic, create a new topic.
findftp
Legendary
*
Offline Offline

Activity: 1008
Merit: 1000


Delusional crypto obsessionist


View Profile
October 23, 2014, 09:10:30 AM
 #21

Nice graphics and nice work!  Wink

What software did you use?


Excel.

As Rpietila says, you can do amazing things with excel.

Wow, nice work!
Thread watched and notified.
Just gave a small tip.

1FdwDuV2qgsw2w1Lza8nsea7L6mQxsc7g3
1571473840
Hero Member
*
Offline Offline

Posts: 1571473840

View Profile Personal Message (Offline)

Ignore
1571473840
Reply with quote  #2

1571473840
Report to moderator
1571473840
Hero Member
*
Offline Offline

Posts: 1571473840

View Profile Personal Message (Offline)

Ignore
1571473840
Reply with quote  #2

1571473840
Report to moderator
1571473840
Hero Member
*
Offline Offline

Posts: 1571473840

View Profile Personal Message (Offline)

Ignore
1571473840
Reply with quote  #2

1571473840
Report to moderator
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
Trolololo
Sr. Member
****
Offline Offline

Activity: 259
Merit: 268



View Profile
October 23, 2014, 09:22:32 AM
 #22

Added to the OP:
Calculate today's trendline price HERE

Your work is really great. This trendline seems realistic.

Could you provide a date for the one million dollar price ? Or extend your chart a little bit ?


1.000.000 on 06-09-2026

I would like extend the chart to that date, but by excel worksheet doesn't want to. I don' know why.
Trolololo
Sr. Member
****
Offline Offline

Activity: 259
Merit: 268



View Profile
October 23, 2014, 09:23:56 AM
 #23

Nice graphics and nice work!  Wink

What software did you use?


Excel.

As Rpietila says, you can do amazing things with excel.

Wow, nice work!
Thread watched and notified.
Just gave a small tip.

Thank you!
madmat
Legendary
*
Offline Offline

Activity: 966
Merit: 1000



View Profile
October 23, 2014, 09:34:24 AM
 #24

Added to the OP:
Calculate today's trendline price HERE

Your work is really great. This trendline seems realistic.

Could you provide a date for the one million dollar price ? Or extend your chart a little bit ?


1.000.000 on 06-09-2026

I would like extend the chart to that date, but by excel worksheet doesn't want to. I don' know why.

Thanks for the date.

Excel is not perfect. Did you reach the line limit ? (1 048 576 lines)
Trolololo
Sr. Member
****
Offline Offline

Activity: 259
Merit: 268



View Profile
October 23, 2014, 11:54:39 AM
 #25

What price do you get in 2075? In my opinion the "equilibrium" price of Bitcoin is one of the easier numbers to make rational arguments about (thermodynamics is always on sounder footing than kinetics, etc.).
For example, if you get $1 trillion per BTC in 2075, I would say the model is problematic, however well fitted. If you get something in the $1-10 million range, then you're probably in the ballpark. (these models assume a certain optimism about adoption already)


With only 1550 days of prices, it's quite weird to estimate the value for day #24.098 (01-01-2075).

But let's suppose that the price stays flat at 370 for 5 more years (from today to 01-01-2020).
That would mean we have around 5500 days of prices.
The logarithmic regression would throw a estimated value of 961 k$ by 01-01-2075.

That's a more realistic value than any linear regression model.
oda.krell
Legendary
*
Offline Offline

Activity: 1470
Merit: 1007



View Profile
October 23, 2014, 12:30:53 PM
 #26

Nice post & analysis, Trolololo. Following.

I made a similar observation earlier this month in Stephen Reed's thread, about what looked like an increasing time span between reaching the next order of magnitude in network size, but didn't follow up on it (shameless plug to my own post :D).


A question.

I'm not sure about the exact type of growth of your function. What you seem to be mapping is logarithmic growth (slow) on a log chart (fast). Stripped of its minor constants, your formula is of the form 10^ln(t) for time = t. ln(t) grows extremely slow over time, but the result is used as a positive, growing exponent.

It seems to fall somewhere between linear growth and exponential growth, and it isn't bounded either (like in Stephen's model). I was wondering if someone with more knowledge on functional growth could answer this once and for all for me, have been wondering about this for a while now. (EDIT: I'm wondering if it could be an instance of so called sub-exponential growth)


A critical remark.

While I personally, intuitively, find a price function with a declining growth rate (like yours) more plausible than the constant growth rate models that have been presented on this forum (i.e. the "loglinear models" you linked to as well), one problem still remains:

Price tends to "jerk around" all those models.

I remember that, late last year, when price exceeded even the loglinear model's predictions, some analysis was posted that suggested a superexponential price function to model BTC price.

Then came the first leg of the 2013/2014 correction, and suddenly the loglinear models were all the rage again.

Now, the correction continues, and you suggest (with good reasons, I agree) a model based on an below exponential growth assumption. But I'm afraid all it takes is another year of bear market (or perhaps, a sudden rally of huge proportions), and we need to re-adjust our assumption for what the "best" growth type for our model is...

Here's what I'm trying to say: I am using technical (i.e. historic price based) methods myself all the time for predicting price on the short term. However, I start to think that, on a long enough time scale, fundamentals govern the price function. So a model like yours (or Stephen's, or rpietila's), that are essentially an extrapolation from an (admittedly well fitted) function on the historic price data might come to its limits.

I am thinking that perhaps the only semi-reliable way to go about mapping the "long term trend" is Peter R.'s way: finding a proxy for network size, and then modeling expected price/mcap as a function of network size.

See for example here: https://bitcointalk.org/index.php?topic=68655.msg9059346#msg9059346

He still makes a number of assumptions (Metcalfe's law, for example), and in a way, his method only shifts the problem (because now we are trying to predict, i.e. extrapolate, network size), but at least his predicted numbers will rarely be so out of tune with reality as the pure time series models can be at times.

Not sure which Bitcoin wallet you should use? Get Electrum!
Electrum is an open-source lightweight client: fast, user friendly, and 100% secure.
Download the source or executables for Windows/OSX/Linux/Android from, and only from, the official Electrum homepage.
Trolololo
Sr. Member
****
Offline Offline

Activity: 259
Merit: 268



View Profile
October 23, 2014, 01:16:50 PM
Last edit: October 23, 2014, 02:59:06 PM by Trolololo
 #27

Nice post & analysis, Trolololo. Following.

I made a similar observation earlier this month in Stephen Reed's thread, about what looked like an increasing time span between reaching the next order of magnitude in network size, but didn't follow up on it (shameless plug to my own post Cheesy).


A question.

I'm not sure about the exact type of growth of your function. What you seem to be mapping is logarithmic growth (slow) on a log chart (fast). Stripped of its minor constants, your formula is of the form 10^ln(t) for time = t. ln(t) grows extremely slow over time, but the result is used as a positive, growing exponent.

It seems to fall somewhere between linear growth and exponential growth, and it isn't bounded either (like in Stephen's model). I was wondering if someone with more knowledge on functional growth could answer this once and for all for me, have been wondering about this for a while now. (EDIT: I'm wondering if it could be an instance of so called sub-exponential growth)


A critical remark.

While I personally, intuitively, find a price function with a declining growth rate (like yours) more plausible than the constant growth rate models that have been presented on this forum (i.e. the "loglinear models" you linked to as well), one problem still remains:

Price tends to "jerk around" all those models.

I remember that, late last year, when price exceeded even the loglinear model's predictions, some analysis was posted that suggested a superexponential price function to model BTC price.

Then came the first leg of the 2013/2014 correction, and suddenly the loglinear models were all the rage again.

Now, the correction continues, and you suggest (with good reasons, I agree) a model based on an below exponential growth assumption. But I'm afraid all it takes is another year of bear market (or perhaps, a sudden rally of huge proportions), and we need to re-adjust our assumption for what the "best" growth type for our model is...

Here's what I'm trying to say: I am using technical (i.e. historic price based) methods myself all the time for predicting price on the short term. However, I start to think that, on a long enough time scale, fundamentals govern the price function. So a model like yours (or Stephen's, or rpietila's), that are essentially an extrapolation from an (admittedly well fitted) function on the historic price data might come to its limits.

I am thinking that perhaps the only semi-reliable way to go about mapping the "long term trend" is Peter R.'s way: finding a proxy for network size, and then modeling expected price/mcap as a function of network size.

See for example here: https://bitcointalk.org/index.php?topic=68655.msg9059346#msg9059346

He still makes a number of assumptions (Metcalfe's law, for example), and in a way, his method only shifts the problem (because now we are trying to predict, i.e. extrapolate, network size), but at least his predicted numbers will rarely be so out of tune with reality as the pure time series models can be at times.


In linear price axis the trend looks exponential:



But is really a subexponential line, a "lowering" exponential trend.

I suspect that in the very long term (reaching year 2075 i.e.) the linear price axis graph would look like an "S" curve (logistic growth). Pending task...

Edit: no "S" look at all. Looks like a "lowering" exponential trend for any timeframe.
To have a "S" look que should use another regression function, like



Relative to volatility, I also suspect that the tops of the bubbles could fit in a similar logarithmic regression. The same for the bottoms of the bubbles. That could give us an estimation of the roofs and floors of next bubbles.
Another pending task...


dnaleor
Legendary
*
Offline Offline

Activity: 1470
Merit: 1000


Want privacy? Use Monero!


View Profile
October 23, 2014, 04:41:31 PM
 #28

nice charts Smiley

will folow your posts from now on

edit: left you a small tip Smiley

Searching for a trusted escrow service? Check my trust. I use a Trezor device for my escrow services, so your coins are safe. Fee: 0.5% or 0.05 BTC, whichever is lower
I support the largest public transparent p2p ledger Bitcoin (16TwXyEmpz7xKHbyVufZECXGFmUH9wHUyW) and the best private fungible digital cash Monero (dnaleor.weuse.cash)
Cetere mi opinias ke Dasho estas detruenda.     |     1Credit - Fair emission, better than BTC > CLCqECaYpCahgKRsXJiVBzgsiz57ZHfr4u
oda.krell
Legendary
*
Offline Offline

Activity: 1470
Merit: 1007



View Profile
October 23, 2014, 04:46:54 PM
 #29


Edit: no "S" look at all. Looks like a "lowering" exponential trend for any timeframe.
To have a "S" look que should use another regression function, like



That's what I said... the 'S' shape (of slipperyslope, e.g.) is bounded growth, i.e. is convergent. Your function doesn't have a finite limit. It just goes to infinity rather slowly Cheesy

Not sure which Bitcoin wallet you should use? Get Electrum!
Electrum is an open-source lightweight client: fast, user friendly, and 100% secure.
Download the source or executables for Windows/OSX/Linux/Android from, and only from, the official Electrum homepage.
dnaleor
Legendary
*
Offline Offline

Activity: 1470
Merit: 1000


Want privacy? Use Monero!


View Profile
October 23, 2014, 04:46:58 PM
 #30

I remember that, late last year, when price exceeded even the loglinear model's predictions, some analysis was posted that suggested a superexponential price function to model BTC price.

that was fun Cheesy
We would be > 600k by now Wink

edit: just posting it for the lulz:

Searching for a trusted escrow service? Check my trust. I use a Trezor device for my escrow services, so your coins are safe. Fee: 0.5% or 0.05 BTC, whichever is lower
I support the largest public transparent p2p ledger Bitcoin (16TwXyEmpz7xKHbyVufZECXGFmUH9wHUyW) and the best private fungible digital cash Monero (dnaleor.weuse.cash)
Cetere mi opinias ke Dasho estas detruenda.     |     1Credit - Fair emission, better than BTC > CLCqECaYpCahgKRsXJiVBzgsiz57ZHfr4u
Capt Drake
Full Member
***
Offline Offline

Activity: 208
Merit: 100


View Profile
October 23, 2014, 07:28:46 PM
 #31

Very nice work man!!

I'll be watching your future updates on this!  Grin
Trolololo
Sr. Member
****
Offline Offline

Activity: 259
Merit: 268



View Profile
October 23, 2014, 11:06:52 PM
 #32

nice charts Smiley

will folow your posts from now on

edit: left you a small tip Smiley

Thank you!
superresistant
Legendary
*
Offline Offline

Activity: 2030
Merit: 1095



View Profile
October 24, 2014, 11:25:41 AM
Last edit: October 24, 2014, 11:45:15 AM by superresistant
 #33

Nice graphics and nice work!  Wink
What software did you use?
Excel.
As Rpietila says, you can do amazing things with excel.

Impressive.
Yes, Excel (or equivalent) can do so much things.

EDIT :
I can only agree with the estimate, 1.000 USD by 24-04-2015
But that's only an estimation.
smooth
Legendary
*
Offline Offline

Activity: 2282
Merit: 1137



View Profile
October 25, 2014, 07:20:14 AM
 #34

I'm not sure about the exact type of growth of your function. What you seem to be mapping is logarithmic growth (slow) on a log chart (fast). Stripped of its minor constants, your formula is of the form 10^ln(t) for time = t. ln(t) grows extremely slow over time, but the result is used as a positive, growing exponent.

10ln(x) = xln(10)
cbeast
Donator
Legendary
*
Offline Offline

Activity: 1736
Merit: 1002

Let's talk governance, lipstick, and pigs.


View Profile
October 25, 2014, 08:10:10 AM
 #35

Let's now see some new analytics using splines and regressive curvilinear functions. I'm so tired of seeing mspaint lines over boring plots. The speculation threads will love these.

Any significantly advanced cryptocurrency is indistinguishable from Ponzi Tulips.
oda.krell
Legendary
*
Offline Offline

Activity: 1470
Merit: 1007



View Profile
October 25, 2014, 09:51:45 AM
 #36

I'm not sure about the exact type of growth of your function. What you seem to be mapping is logarithmic growth (slow) on a log chart (fast). Stripped of its minor constants, your formula is of the form 10^ln(t) for time = t. ln(t) grows extremely slow over time, but the result is used as a positive, growing exponent.

10ln(x) = xln(10)



Thanks. Feel like an idiot for missing that.

Which would make it about quadratic growth. Interesting.

Not sure which Bitcoin wallet you should use? Get Electrum!
Electrum is an open-source lightweight client: fast, user friendly, and 100% secure.
Download the source or executables for Windows/OSX/Linux/Android from, and only from, the official Electrum homepage.
laurentmt
Sr. Member
****
Offline Offline

Activity: 386
Merit: 250


View Profile
October 25, 2014, 09:10:07 PM
 #37

In what year does your regression intersect with $1M USD?
I would say for Xmas 2026  Wink
smooth
Legendary
*
Offline Offline

Activity: 2282
Merit: 1137



View Profile
October 25, 2014, 09:43:10 PM
 #38

I'm not sure about the exact type of growth of your function. What you seem to be mapping is logarithmic growth (slow) on a log chart (fast). Stripped of its minor constants, your formula is of the form 10^ln(t) for time = t. ln(t) grows extremely slow over time, but the result is used as a positive, growing exponent.

10ln(x) = xln(10)



Thanks. Feel like an idiot for missing that.

Which would make it about quadratic growth. Interesting.

Yes I thought so too. If you buy the Metcalf's Law valuation argument (I don't entirely), that corresponds roughly with linear growth of usage. Which interestingly is what Chris Dixon has said they are seeing in his businesses (mostly Coinbase I think).

macsga
Legendary
*
Offline Offline

Activity: 1484
Merit: 1002


Strange, yet attractive.


View Profile
November 04, 2014, 11:14:28 PM
 #39

Following. Great job indeed. Congrats. Smiley

Chaos could be a form of intelligence we cannot yet understand its complexity.
ssmc2
Legendary
*
Offline Offline

Activity: 1750
Merit: 1033


View Profile
November 04, 2014, 11:20:18 PM
 #40

This is in my weekly repertoire of threads I look at to give me that warm blanket feeling  Grin 
Pages: « 1 [2] 3 4 5 6 7 8 9 10 11 12 13 »  All
  Print  
 
Jump to:  

Sponsored by , a Bitcoin-accepting VPN.
Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!